(A-189) Studying When to Add Biological Therapy to Cytotoxic Therapy in Advanced Cancer Care Using Real World Data (RWD): Challenges and Solutions to Avoid Biases that Generate Immortal Time

Background: Because RWD lack information on treatment assignment (usually the only information available is treatment prescription or administration) and early outcomes can happen in advanced cancer, some design choices may introduce immortal time due to selection or due to treatment misclassification.

Objectives: To illustrate two design choices that would biasedly introduce immortal time and to emulate a target trial that would avoid such bias using the GEMCAD 1401 registry.

Methods: We specified a target trial to estimate the effect of “initiating monoclonal antibodies (mAB) within 8 weeks (grace period) of starting first line chemotherapy (CT)” versus “deferring their addition to second line CT” on the overall survival of patients with advanced colorectal cancer.

The first design choice that would introduce immortal time would be imposing a post-baseline selection by requiring patients to start second line CT. This was avoided by aligning eligibility with time zero (initiation of first line CT).

The second design choice that would introduce immortal time would be misclassifying patients into treatment strategies in the grace period by excluding from the first line mAB arm those who die in the grace period without starting mAB. This was avoided with two methods. First, via clone, censoring and weighting (CCW). Second, by reformulating the first line mAB strategy to “initiating mAB concomitantly with first line CT” so that it was distinguishable at baseline and emulating 8 trials, starting at each of the first 8 weeks.

Survival curves adjusted for baseline and post-baseline covariates were estimated via weighted pooled logistic regression.

Results: There were 1014 eligible patients in GEMCAD (2014-2018). The CCW approach estimated a 4-year survival difference of 4.3% (-3.3% ; 10.0%) (reference: first line mAB ) and a risk ratio (RR) of 0.95 (0.88 ; 1.04), while sequential trial emulation estimated a survival difference of -2.8% (-7.5% ; 1.2%) and a RR of 1.04 (0.98 ; 1.10). Biased analyses that introduced immortal time by selecting and misclassifying patients estimated an implausible early survival benefit (6-months RR of 0.13 (0.03 ; 0.28)). Results of sensitivity analyses were robust to grace period length, weight truncation, distribution of mAB initiation over grace period (CCW design), and length of exposure intervals (sequential emulation design).

Conclusions: Our results are compatible with no effect on survival by the sequencing of mAB. Aligning time zero and treatment assignment with eligibility avoided introducing immortal time due to selection. Using CCW when treatment strategies are not distinguishable at baseline or changing the research question to make them distinguishable at baseline avoided introducing immortal time due to treatment misclassification.